This thesis investigates learning-based approaches for fault localization in electric power grids using high-resolution voltage and current measurements. Accurate identification of fault locations is essential for maintaining system reliability and enabling fast restoration after disturbances. Traditional analytical methods rely on simplified physical models, which can lead to reduced accuracy under varying operating conditions.
The goal of this work is to explore hybrid approaches that combine established physical knowledge from power system analysis with modern machine learning techniques. By integrating known operators with data-driven models, the thesis aims to study whether such hybrid methods can improve the robustness and accuracy of fault localization.
The developed approaches will be evaluated using simulated power system data representing various fault scenarios and operating conditions. The results will provide insights into the potential of hybrid learning methods for supporting future intelligent protection and monitoring systems in electric power grids.